Performance Comparison with Only Top 20 detail features

• For the detailed feature type, to find how the most important features for each gold standard set by RF-Gini Importance perform
• To see the performance dropping when only use the top N important features. Here we use N = 20.

• Derived performance for each gold standard based on each’ top 20 features within the detail style

• Precision vs. recall curves comparisons: through Random Forest (RF) - 200 tress: ( Top20 features vs. Full-Detail 162 features )
• è The top 20 detail features did a pretty good job compared to the full 162 detail feature set

---

1. Co-Complex Protein Pair Prediction Summary

• The Top 20 features ordering in the training/test sets (MIPS)

• 3 4 5 8 9 13 14 16 19 20 23 39 48 49 61 76 85 100 101 105

• The Top 20 features for the MIPS (co-complex task):

	Feature Index	Description	Note:	The RF-Imp + Top 20 Feature Ranked list
1	3	Sporulation	Gene exp	0.0059 61 0.0061 85 0.0068 100 0.0070 105 0.0090 14 0.0095 23 0.0100 48 0.0139 20 0.0228 4 0.0241 3 0.0385 9 0.0430 5 0.0441 13 0.0569 16 0.0715 8 0.0731 101 0.0778 19 0.0965 39 0.1182 76 0.1637 49
2	4	fermentation to respiration / cell wall	Gene exp
3	5	Eisen1998 stress	Gene exp
4	8	Jelinsky2000 alkylating agents	Gene exp
5	9	Lyons2000 zinc	Gene exp
6	13	Shamji2000	Gene exp
7	14	Travers2000	Gene exp
8	16	Gasch2001	Gene exp
9	19	Gasch2000 stress	Gene exp
10	20	cell cycles	Gene exp
11	23	transcription regulator activity	Go Func
12	39	structural molecule activity	Go Func
13	48	protein catabolism	Go Proc
14	49	protein biosynthesis	Go Proc
15	61	Transcription	Go Proc
16	76	Ribosome	Go Comp
17	85	mitochondrial membrane	Go Comp
18	100	HMS_PCI Mass
19	101	TAP Mass
20	105	Sequence Similarity

---

2. Direct Protein-Protein Interaction Prediction Summary

• The Top 20 features ordering in the training/test sets (DIPS)

• 3 5 8 20 23 30 48 61 67 74 83 100 101 102 103 105 109 110 157 162

• The Top 20 features for DIPS physical binding task

	Feature Index	Description	Note:	The RF-Imp + Top 20 Feature Ranked list
1	3	sporulation	Gen Exp	0.0130 109 0.0140 162 0.0140 30 0.0141 74 0.0153 48 0.0171 23 0.0172 61 0.0189 157 0.0200 3 0.0209 103 0.0234 20 0.0261 83 0.0267 8 0.0306 105 0.0336 5 0.0352 110 0.0372 67 0.0459 100 0.0611 102 0.1035 101
2	5	Eisen1998 stress	Gen Exp
3	8	Jelinsky2000 alkylating agents	Gen Exp
4	20	cell cycle ( 76 col )	Gen Exp
5	23	transcription regulator activity	Go Func
6	30	RNA binding	Go Func
7	48	protein catabolism	Go Proc
8	61	transcription	Go Proc
9	67	nucleus	Go Proc
10	74	Golgi apparatus	Go Proc
11	83	vesicle-mediated transport	Go Comp
12	100	HMS_PCI Mass	HMS_PCI Mass
13	101	TAP Mass	TAP Mass
14	102	Y2H	Y2H
15	103	Synthetic Lethal	Synthetic Lethal
16	105	Celegan homology PPI	Sequence Similarity
17	109	celegan	Homology
18	110	Domain-Domain Interaction	Domain-Domain Interaction
19	157	Cell cycle defects	MIPS protein phenotype
20	162	Nucleic acid metabolism defects	MIPS protein phenotype

---

3. Co-pathway Protein Pair Prediction Summary

• The Top 20 features ordering in the training/test sets (KEGG04)

• 3 4 5 8 9 13 14 16 19 20 24 28 39 41 49 51 76 98 104 157

• The Top 20 features for keGG04

	Feature Index	Description	Note:	The RF-Imp + Top 20 Feature Ranked list
1	3	sporulation	Gen Exp	0.0118 41 0.0127 98 0.0142 24 0.0142 28 0.0147 14 0.0170 20 0.0204 3 0.0207 104 0.0254 13 0.0255 4 0.0310 157 0.0356 51 0.0474 9 0.0511 39 0.0571 8 0.0584 16 0.0726 19 0.0729 5 0.0757 76 0.0867 49
2	4	fermentation to respiration / cell wall	Gen Exp
3	5	Eisen1998 stress	Gen Exp
4	8	Jelinsky2000 alkylating agents	Gen Exp
5	9	Lyons2000 zinc	Gen Exp
6	13	Shamji2000	Gen Exp
7	14	Travers2000	Gen Exp
8	16	Gasch2001	Gen Exp
9	19	Gasch2000 stress	Gen Exp
10	20	cell cycle	Gen Exp
11	24	transferase activity	Go Func
12	28	oxidoreductase activity	Go Func
13	39	structural molecule activity	Go Func
14	41	nucleotidyltransferase activity	Go Func
15	49	protein biosynthesis	Go Proc
16	51	carbohydrate metabolism	Go Proc
17	76	ribosome	Go Comp
18	98	Prot Exp	Prot Exp
19	104	Gene Neigh/Fusion/Cocur	Gene Neigh/Fusion/Cocur
20	157	Cell cycle defects	MIPS phenotype